US 7590063 B2 Abstract A technique for binning aperiodic latency sample data using a data representation called latency band graphs. A fluid flow analysis produces a small, fixed size set of automatically generated bins dependent only on the timeline defined by periodic traffic. The compact number of bins yields a parameterized latency representation suitable for real-time estimation and goodness-of-fit tests.
Claims(11) 1. A method for managing data flow in a system with simultaneous scheduling of aperiodic messages and periodic transmissions on a common bus, comprising the steps of:
(a) using predefined periodic transmission times, calculating data transition points between periodic and aperiodic message transmissions intervals for a hyperperiod of interest in said system;
(b) using said data transition points to produce a series of aperiodic latency estimation inflection points;
(c) collecting data points of aperiodic message transmissions for the hyperperiod of interest in said system;
(d) estimating the aperiodic latency probability at an inflection point in the hyperperiod of interest as being equal to the number of sample data points less than or equal to the said inflection point divided by the total number of collected aperiodic latency sample data points, said data points forming a data point plot that is assumed to be linear between said aperiodic latency inflection points; and
(e) transmitting the aperiodic messages based, at least in part, on the estimated aperiodic latency probability.
2. The method of
3. The method of
4. The method of
5. A method for managing data flow in a system with simultaneous scheduling of aperiodic messages and periodic transmissions on a common bus, wherein data points of aperiodic message transmissions for the hyperperiod of interest in said system are collected, the method comprising:
using predefined periodic transmission times to calculate data transition points between busy and idle intervals for a hyperperiod of interest in said system;
using said data transition points to produce a series of aperiodic latency estimation inflection points;
estimating the aperiodic latency probability at an inflection point in the hyperperiod of interest as being equal to the number of sample data points less than or equal to the said inflection point divided by the total number of collected aperiodic latency sample data points, said data points forming a data point plot that is assumed to be linear between said aperiodic latency inflection points; and
transmitting the aperiodic messages based, at least in part, on the estimated aperiodic latency probability.
6. The method of
7. The method of
8. The method of
9. A system with simultaneous scheduling of aperiodic messages and periodic transmissions, the system comprising:
a bus on which the periodic and aperiodic messages are transmitted; and
a processor operable to use predefined periodic transmission times to calculate points indicating transitions between busy and idle intervals for a hyperperiod of interest in said system; the processor further operable to use said calculated points to produce a series of aperiodic latency estimation inflection points and to estimate the aperiodic latency probability at an inflection point in the hyperperiod of interest as being equal to the number of sample data points less than or equal to the said inflection point divided by the total number of collected aperiodic latency sample data points, said data points forming a data point plot that is assumed to be linear between said aperiodic latency estimation inflection points;
wherein the processor transmits the aperiodic messages on the bus based, at least in part, on the estimated aperiodic latency probability.
10. The system of
11. The system of
Description This invention was made with Government support under the terms of contract DAAH01-00-C-R226 awarded by U.S. Army Redstone. The Government has certain rights in the invention. The present invention relates to simultaneous scheduling random or aperiodic messages and periodic transmissions on common hardware. More particularly, the invention relates to transmission of periodic transmissions that are regarded as flight critical and have predefined, static transmission times, whereas the non-critical applications are transmitted in the time remaining. Distribution control systems must support both periodic functions and aperiodic functions that send and receive data over one or more common data buses. Sending messages on a common data bus or scheduling tasks on a common processor are canonical examples. To ensure guaranteed latencies, that is, network delays, for closed loop periodic control functions such as sensor read, control, actuator write, and the like, static scheduling techniques are presently used to produce a timeline where known data are transmitted at predefined times, leaving a sequence of gaps for aperiodic message transmission. There are currently no scheduling models that allow efficient platforms to be built that predictably support both types of applications. For critical aperiodic functions, such as event-triggered ones, with deadline requirements, such as pilot input, target tracking, alarm signals and the like, sufficient bandwidth to allow for worst case event arrival is statically reserved. This results in an over-engineered system and recurring hardware costs. For non-critical aperiodic functions such as internet connections, voice or video, for example, system design has heretofore been by trial-and-error. There has been no way to predict aperiodic message latencies. The very broad concept of latency measurements in a communication system is taught Link et al. U.S. Pat. No. 6,012,096 relates to a method for network latency in which data packets are transmitted between the users. Skurdal et al. U.S. Pat. No. 6,161,009 uses a control circuit to turn a transmitter on and off, measuring the latency time and using the data in subsequent transmission of data. Dean et al. U.S. Pat. No. 6,332,178 discloses a method for estimating statistics of properties of transactions processed by a memory sub-system of a computer system. This broadly discloses statistical sampling. Goldberg U.S. Pat. No. 5,767,785 discloses a different method for generating prediction recommendations for signals. Grochowski et al. U.S. Pat. No. 6,035,389 teaches a latency vector system using a register latency table. Borella et al U.S. Pat. No. 6,182,125 discloses a determined network latency method. Hershey et al. U.S. Pat. No. 5,375,070, Chuah U.S. Pat. No. 6,115,390, Black et al. U.S. Pat. No. 6,038,599, Foore et al. Publication 2001/002119, and Skene et al. Publication 2001/0052016 are other patents that are generally related to network latency. The general problem of estimating aperiodic latencies in complex systems is enormously difficult, especially for multi-class traffic, which, without special restrictions, is not well understood. A first area of concern are those in which steady state approximations can be made, without estimations for transient behavior. A mathematical analysis of aperiodic message latency distributions is rarely tractable, so it must be estimated using empirical data. For estimating latency distributions, one approach is to construct an empirical distribution function (EDF) using (simulation) data. Another approach is to construct an (approximate) analytic model which would be solved numerically. In the former, PDF's are constructed using a sample of k independent and identically distributed (iid) latency values, {x It would be of great advantage in the art if a greater understanding of steady state approximations of latencies could be achieved. It would be another great advance in the art if a sampling point strategy could be devised that would produce a compact estimate of aperiodic latency distribution. Other advantages will appear hereinafter. It has now been discovered that the above and other objects of the present invention may be accomplished in the following manner. Specifically, the present invention includes a new technique for binning aperiodic latency sample data, or defining cluster regions for latency data. The invention employs latency band graphs to which fluid flow analysis is applied, leading to a small, fixed set of automatically generated bins that depends only on the periodic message traffic. The small and deterministic number of bins yields a compact latency estimation representation suitable for use in real-time on actual data, and provides good latency estimates under a broad range of conditions. It has been discovered that patterns in the size and spacing of gaps in the periodic timeline introduce event latency bands that can have significant effects on the shape of the latency distribution. For a more complete understanding of the invention, reference is hereby made to the drawings, in which: The present invention is admirable suited to manage data to ensure guaranteed deadlines for critical closed loop periodic control functions and other related functions. In order to describe the invention, the following definitions are used.
The m periodic message durations have blocking times m The interarrival times between adjacent aperiodic message arrives are iid with distribution A, mean λ In its simplest form, the invention generates a set of points or bins, collects and categorizes a large number actual data points, and places the categorized points in the generated bins. Thus, instead of 500 or more data points, the present invention considers only a small number, often less than 12. It is helpful to use a single simple reference model to provide qualitative and intuitive comparisons between observed system behaviors and expected behaviors for a proportionally shared server. Aperiodic traffic streams can have either exponential interarrival and service times, as in the equation α The latency distribution graphs shown in The latency band graph of For a given periodic timeline BI, latency bands are constructed as follows. The x-axis runs from 0 to H and the y-axis run from 0 to d Aperiodic work is discharged in idle intervals if it backlogs and accumulates in periodic busy intervals. The flow dynamics of the aperiodic message queue are likely to change and busy/idle transition points. These transition points are likely to correspond to points of inflection in the latency distribution. Inclusion of points in BI is the first step in generating binning points. For a given latency range, the band structure gives some insight into arrival times. There are confounding arrival conditions for aperiodic latencies among neighboring bands. Even though the arrival times are not controlled, when conditioning on arrival time, the set of possible latency values changes. For example, if an aperiodic message latency falls between 52 and 64, then either the arrival time of the message occurred in the interval [0, 12] or [48, 64], and at no other time within the hyperperiod. Intervals [0, 12] and [48, 64] are band confounding regions for latencies in the range [52, 64]. Approximating confounding regions for bands is the second phase of bin point generation. In the latency band graph, the busy/idle interval backlog/discharge is seen along the y-axis as a function of the arrival time modulo the hyperperiod of a queued message. Points identifying confounding region boundaries might be points of inflection in a latency distribution, since within the confounding regions there are multiple conditions leading to the observed dynamics, at least one of which changes when exiting the region. Bin generation consists of two phases. Let S be a set of binning points. Initialize S={b Further binning points are needed to group latencies in bands with confounding regions. Computing these separating points is done with a set of linear equations. Initialize S=BI={0.75H, H}. For each point s Let S={s: sεS}∪{H}. In Then let x Finally, define By picking BI to contain gap and block durations that are multiples of one another, one can attempt to reduce the final number of binning points. Points in S that are close to one another might be collapsed, either by deleting one or using an average. When n Table 2 shows values for BI the busy/idle transition points in a hyperperiod H. ρ
In On the left hand side of For latency bands with blocking, point p When L
Only when the hyperperiod is suitably long does the fluid flow discharge behavior shown in When generating the final set of binning points, the values of ρ Including additional binning points at the y-values in p When system utilization ρ is low, most all aperiodic message arrivals will have a minimal wait. The right hand side of Unlike when blocking occurs in an aperiodic transmission band, once the aperiodic backlog is discharged it remains in the discharged state until the next periodic blocking band begins. When the periodic blocking band with duration block The data shown in the figures has been taken from several dozen simulations that represent a spectrum of operational settings. At one end of the spectrum, the hyperperiods are very long relative to local gaps and blocks. At the other end of the spectrum, the cumulative gap time available in any hyperperiod is small relative to the pending aperiodic work, in which case latency distributions are reasonably well approximated by the equation defining M/M/1 aperiodic queue simulation data. In summary, the invention comprises the generation of a set of binning points S=(s The latency bands in In The latency bands corresponding to the latencies in There are several benefits from the present invention. The number of bins is compact and deterministic. The number of support points, x-values with non-zero probability, is defined by the number of bins, n, not by the number of sample points k, where often n<<k. In the figures shown herein, k is always 500 and n Because the number of bins n depends only on the timeline generated by periodic message transmission, the support points are known so latency values observed on-line can be quickly recorded. There is considerably greater variability in recording times when k is the sample size, compared to when n is used. While particular embodiments of the present invention have been illustrated and described, it is not intended to limit the invention, except as defined by the following claims. Patent Citations
Non-Patent Citations
Referenced by
Classifications
Legal Events
Rotate |